ImportError: libmkl_intel_lp64.so.1

[adammz1]

Faiss version: 1.7.4

Installed from: anaconda

Running on:

• GPU

Interface:

• Python

I followed the installation instructions in faiss/INSTALL.md:

conda create --name faiss_1.7.4 python=3.10
conda activate faiss_1.7.4
conda install faiss-gpu pytorch pytorch-cuda -c pytorch -c nvidia

Then I ran a python3.10 script containing "import faiss" and received error:

"from . import _swigfaiss
ImportError: libmkl_intel_lp64.so.1: cannot open shared object file: No such file or directory"

[adammz1]

I built from source instead-
Faiss 1.7.4 did not work- extremely long delay when moving to device. I downgraded to Faiss 1.7.3 and it worked. System info:
OS: Ubuntu 22.04, CUDA 11.7.0, Python 3.10, NVDIA RTX 3080

[algoriddle]

This is a bug in Faiss 1.7.4, that the conda package doesn't specify the dependencies correctly. The fix is to install mkl separately, conda install mkl=2021. This is already fixed in the nightly.